TAPL: Dynamic Part-based Visual Tracking via Attention-guided Part Localization
This work addresses visual tracking challenges for applications requiring robustness to appearance variations, but it appears incremental as it builds on part-based methods with attention mechanisms.
The authors tackled the problem of visual tracking performance drop under large appearance changes like deformation and occlusion by proposing a dynamic part-based tracker that updates target part representations and uses an attention-guided network for part localization, achieving promising results on benchmarks such as VOT2018, OTB100, and GOT-10k.
Holistic object representation-based trackers suffer from performance drop under large appearance change such as deformation and occlusion. In this work, we propose a dynamic part-based tracker and constantly update the target part representation to adapt to object appearance change. Moreover, we design an attention-guided part localization network to directly predict the target part locations, and determine the final bounding box with the distribution of target parts. Our proposed tracker achieves promising results on various benchmarks: VOT2018, OTB100 and GOT-10k